A cmux alternative that runs Claude Code and Codex on whole projects.

Duplicate turns the count into projects instead. Pick worktrees or standalone copies, set how many — up to 50 — choose the action or command each one runs, and type the prompt once. Every copy inherits the project's services and actions, and the sidebar shows what each agent is doing.

None of this isolates a port or a database. What lpm does is check every declared port as the copy comes up, and name whatever is already sitting on one — a warning at the door, not a partition. See the isolation models compared.

Not much to convert — the two files describe different things. cmux.json configures your terminal; .lpm.yml describes your stack.

// app-wide: shortcuts, sidebar, notifications,
// actions, custom commands, workspace layouts
// (.cmux/cmux.json adds per-repo actions)

services:
  api:
    cmd: npm run dev
    port: 3000
  worker:
    cmd: npm run worker

profiles:
  light: [api]

• cmux list-workspaces→lpm list --json
• cmux notify→lpm set-status

  A badge on the tab, a chime, a macOS banner when you are away from the window, and a push to your phone.

• socket control→~/.lpm/lpm.sock

  One shell-quoted command per line.

• a tab per agent→a project copy per agent (lpm worktree / lpm duplicate)
• a cmux action or custom command→an action in .lpm.yml — a button, or lpm run

  Declared once per project, so every copy of that project has it too.

Anything in the right-hand column that changes state — lpm start, lpm worktree, lpm service web restart, lpm run — is routed through lpm itself, so leave the app open when a script calls one. lpm list and lpm logs read the running services themselves, so they answer whether lpm is open or closed. The exception among the readers is lpm status, which prints what your agents are doing: the app is where those states are kept.

Keeping both is normal: cmux stays your terminal, lpm decides which project is up.

• What is cmux?

  A native macOS terminal from Manaflow built around AI coding agents: vertical tabs showing branch and PR status, split panes, a notification panel, a scriptable browser pane, and a CLI plus Unix socket to control all of it. It renders through libghostty and reads your Ghostty config for themes and fonts.
• Is cmux free and open source?

  Yes. cmux ships under GPL-3.0-or-later; its repository adds that commercial terms may be available where the GPL will not do, and that a paid Founder's Edition buys early access to features still in progress. lpm is MIT and has nothing to buy. Cost is not the reason to choose between them.
• Can I run several Claude Code or Codex agents at once?

  In cmux each agent gets its own tab, so two agents in the same folder still edit the same files. lpm splits the repo first — up to 50 copies at a time, each one a linked worktree on a branch of its own or a full folder copy that keeps its Git history — and queues the same prompt in every copy, with working, needs you, done or a problem on the agent tab that owns it. What the copies still share is ports and databases.
• Can lpm and cmux run side by side?

  Yes, and it is a reasonable setup. cmux configures your terminal; lpm describes your projects. Neither reads the other's config.
• How do I move a cmux setup to lpm?

  There is nothing to convert: cmux.json describes your terminal, not your stack. Add the folder in lpm, then hit Generate with AI in its config editor — the Claude Code or Codex you already have reads the repo, compose file included, and drafts the services for you to prune. A command you kept as a cmux action becomes an lpm action: one click, or lpm run.
• Does lpm need tmux?

  No — and cmux does not need it either. lpm never puts your services inside tmux and does not require it on the machine: the services keep running when you quit lpm, reopening the app picks them up again, and there is no .tmux.conf anywhere in that.